PROJECT SUMMARY About 2/3 of cancer patients in US receive radiation therapy either alone or in conjunction with surgery, chemotherapy, immunotherapy, etc. Treatment planning, where an optimal treatment strategy is designed for each individual patient and executed for the whole treatment course, is analogous to the design of a blueprint for building construction. If a treatment plan is poorly designed, the desired treatment outcome cannot be achieved, no matter how well other components of radiation therapy are performed. In the current clinical workflow, a treatment planner works towards a good quality plan in a trial-and-error fashion. Many rounds of consultation between the planner and physician are needed to reach a plan of physician's satisfaction, because physician's preference for a particular patient can hardly be quantified and precisely conveyed to the planner. Consequently, planning time can be up to a week for complex cases and plan quality may be poor and can vary significantly due to varying levels of physician and planner's skills and physician-planner cooperation, etc., which substantially deteriorates treatment outcomes. For example, head and neck (H&N) cancer patients treated with suboptimal plans present 20% lower 2-year overall survival and 24% higher 2-year local-regional failure. Prolonged overall treatment process due to treatment planning reduces local-regional control rate by 12?14% per week. Furthermore, as patient's anatomy can rapidly change within the planning time, the optimally designed plan becomes inappropriate for the changed anatomy. Recently, artificial intelligence (AI) has made colossal advancements. We believe that AI technologies have a great potential to revolutionize treatment planning. Treatment planning consists of two major aspects: commonality and individuality. By exploiting the commonality through deep supervised learning, we can develop a treatment plan as good as those for previously treated similar patients. The individuality can be actualized by learning physician's special considerations for a particular patient using deep reinforcement learning. Our preliminary studies have demonstrated feasibility of these ideas. We hypothesize that an AI-based intelligent treatment planning system can consistently produce high-quality treatment plans with extremely high efficiency. This hypothesis will be tested using H&N cancer patients as a test bed via two aims. Aim 1, System development. Develop two deep-learning models to realize the proposed treatment planning workflow and incorporate them into a clinical environment. Aim 2, System validation. Acquire and analyze planning data before and after system implementation. The innovation of this project is the use and customization of the state-of-the-art AI techniques to solve a clinically important problem. These technologies would revolutionize treatment planning process, leading to the efficient generation of consistently high quality plans, irrespective of human skills, experiences, and communications, etc. Besides the significance demonstrated for the H&N cancer patients, the system can be easily extended to other tumor sites, yielding more substantial impacts.